Instructions to use 3dcf-labs/ProximaA-2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use 3dcf-labs/ProximaA-2 with Transformers:

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("3dcf-labs/ProximaA-2", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use 3dcf-labs/ProximaA-2 with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf 3dcf-labs/ProximaA-2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf 3dcf-labs/ProximaA-2:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf 3dcf-labs/ProximaA-2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf 3dcf-labs/ProximaA-2:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf 3dcf-labs/ProximaA-2:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf 3dcf-labs/ProximaA-2:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf 3dcf-labs/ProximaA-2:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf 3dcf-labs/ProximaA-2:Q4_K_M

Use Docker

docker model run hf.co/3dcf-labs/ProximaA-2:Q4_K_M

LM Studio
Jan
Ollama
How to use 3dcf-labs/ProximaA-2 with Ollama:
```
ollama run hf.co/3dcf-labs/ProximaA-2:Q4_K_M
```
Unsloth Desktop

How to use 3dcf-labs/ProximaA-2 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf 3dcf-labs/ProximaA-2:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "3dcf-labs/ProximaA-2:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use 3dcf-labs/ProximaA-2 with Docker Model Runner:
```
docker model run hf.co/3dcf-labs/ProximaA-2:Q4_K_M
```

Lemonade

How to use 3dcf-labs/ProximaA-2 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull 3dcf-labs/ProximaA-2:Q4_K_M

Run and chat with the model

lemonade run user.ProximaA-2-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use 3dcf-labs/ProximaA-2 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf 3dcf-labs/ProximaA-2:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default 3dcf-labs/ProximaA-2:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use 3dcf-labs/ProximaA-2 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf 3dcf-labs/ProximaA-2:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "3dcf-labs/ProximaA-2:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

ProximaA 2 Flagship

ProximaA 2 Flagship is a 27B cybersecurity and software engineering model built by Yevhenii Molchanov at 3DCF Labs. This is a final release. The exact files listed below passed the basic release gate. Broader testing and data review continue, so the full development plan is not yet marked complete.

The model is intended for legitimate security research, authorized pentesting, secure code review, reverse engineering, forensics, incident response and remediation. It is tuned to avoid blanket refusals after an authorized scope is established. The optional Proxima broker can enforce target scope, expiry and confirmation for destructive actions.

In this release, uncensored means that legitimate, in-scope security work should not be blocked by a broad content filter. It does not grant permission to access any system, and it does not make unverified execution claims valid.

Exact release identity

Base: Qwen/Qwen3.8-27B@1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0
Merged-tree manifest SHA-256: eec042f4afcadbb45e7f281b1ce75a2adcc385c175640258ee7cd8dbaf02d36c
Q4_K_M SHA-256: f69e5481bb10b4d2fe2a60b15bfb91f6f650c4aa6fe2a3484af51b8d7d1bddad
Source adapter SHA-256: 00a117b77a009c6a2979e924d5a8196797bd36691de44151c639597f1b1acf21
Candidate manifest SHA-256: 7d7509f464495a79a6a44ae15a58eef28e3351495181aee84485f0c9fc2242f0
Basic gate SHA-256: 7f36ee36802f01c06cdd845f747096dc93b62d7c38ec5531028b8b6504489590
Accelerated release eligible: true
Full development-plan validation complete: false

Verified basic checks

Broker tool use in thinking and non-thinking modes: pass
Coding and security behavior smoke tests: pass
Contained pentesting subset with exact flags: pass
Exact adapter activation: pass
Release identity: pass
Low-refusal authorized-security behavior: pass
Private ProximaA-1 comparison: pass
Q4 hash and smoke test: pass

Detailed test results and their scope are published in the 3DCF Labs GitHub repository. Claims about commands, execution or findings need tool evidence.

Runtime profiles

Use the versioned configs/prompt-profiles.json profiles from the 3DCF Labs GitHub repository. The default identity is ProximaA 2 Flagship. Both thinking-on and thinking-off modes were exercised by the authenticated Q4 gate.

Formats

model/: merged BF16 source checkpoint
gguf/: Q4_K_M build
adapter/: source LoRA adapter

Responsibility

ProximaA is provided AS IS, without warranties or guarantees. It is for experiments, research and authorized work. Use it only on systems you own or have explicit permission to assess.

You are solely responsible for how you use or integrate the model. This includes permissions, targets, commands, outputs, validation, compliance and consequences. To the maximum extent permitted by law, 3DCF Labs, its authors, contributors, licensors and affiliates accept no responsibility or liability for any use, misuse, action, decision, loss, claim or damage involving ProximaA.

Downloads last month: 67

GGUF

Model size

27B params

Architecture

qwen35

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for 3dcf-labs/ProximaA-2

Base model

Qwen/Qwen3.8-27B

Quantized

(1051)

this model